List your company

Multi-turn math reasoning on SHARDED

82.4Accuracy+ FiC
1.2822.3443.464.46May 23, 2026

Evaluation results

MethodLinks
+ FiCModel=PHI-4-14B2026.0582.4
+ FiCModel=QWEN2.5-7B2026.0578.7
+ FiCModel=OLMO-2-13B2026.0577.1
+ FiCModel=LLAMA-3.1-8B2026.0571.7
+ PROMPT-SELFCHECKModel=PHI-4-14B2026.0568.2
+ PROMPT-RULEModel=PHI-4-14B2026.0567.9
+ GATED-SELFModel=PHI-4-14B2026.0563
+ PROMPT-RULEModel=OLMO-2-13B2026.0561.3
SHARDED (no intervention)Model=PHI-4-14B2026.0561.2
+ FiCModel=LLAMA-3.2-3B2026.0560.4
SHARDED (no intervention)Model=QWEN2.5-7B2026.0558.9
+ PROMPT-SELFCHECKModel=OLMO-2-13B2026.0556.1
SHARDED (no intervention)Model=OLMO-2-13B2026.0554.4
+ PROMPT-RULEModel=QWEN2.5-7B2026.0554.1
+ PROMPT-SELFCHECKModel=QWEN2.5-7B2026.0551.6
+ GATED-EXTERNAL (4o-mini)Model=QWEN2.5-7B2026.0545.6
SHARDED (no intervention)Model=LLAMA-3.1-8B2026.0544.7
+ GATED-EXTERNAL (4o-mini)Model=PHI-4-14B2026.0542.5
+ PROMPT-SELFCHECKModel=LLAMA-3.1-8B2026.0539.9
SHARDED (no intervention)Model=LLAMA-3.2-3B2026.0539.2
+ GATED-EXTERNAL (4o-mini)Model=OLMO-2-13B2026.0537.9
+ GATED-SELFModel=QWEN2.5-7B2026.0537.7
+ GATED-EXTERNAL (4o-mini)Model=LLAMA-3.1-8B2026.0536.8
+ PROMPT-RULEModel=LLAMA-3.1-8B2026.0531.3
+ GATED-SELFModel=LLAMA-3.2-3B2026.0530.7
+ GATED-SELFModel=OLMO-2-13B2026.0530
+ PROMPT-RULEModel=LLAMA-3.2-3B2026.0529.4
+ GATED-EXTERNAL (4o-mini)Model=LLAMA-3.2-3B2026.0528.5
+ PROMPT-SELFCHECKModel=LLAMA-3.2-3B2026.0527.2
+ GATED-SELFModel=LLAMA-3.1-8B2026.054.4